Foreign object segmentation in chest x-rays through anatomy-guided shape insertion

2025-01-21

AI Summary

Foreign objects in chest X-rays—such as pacemakers, stents, and ingested items—exhibit high morphological variability, suffer from costly manual annotation, and are severely underrepresented in real clinical datasets, critically limiting instance segmentation performance. To address this data scarcity, we propose an anatomy-guided synthetic data generation paradigm: geometric shapes are synthesized based on anatomical masks and inserted via cut-paste; contrast and opacity are randomized, and semi-automatic label extraction ensures pixel-accurate ground truth. Leveraging only 7% human annotations, our method produces high-fidelity synthetic training data. Fine-tuning Mask R-CNN on this data achieves segmentation accuracy comparable to fully supervised baselines. Crucially, this is the first approach to deeply integrate anatomical priors into the synthetic data generation pipeline—explicitly modeling organ-level spatial constraints—thereby establishing a generalizable, low-resource framework for medical image segmentation. The paradigm significantly reduces annotation burden and lowers barriers to clinical deployment.

Technology Category

• Computer Vision: Segmentation
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Semantics and Knowledge: Representation, semantic annotation, enhancement, enrichments, access and/or integration of a variety of data on the Web
• Security and Privacy: Data transparency and provenance

Abstract

In this paper, we tackle the challenge of instance segmentation for foreign objects in chest radiographs, commonly seen in postoperative follow-ups with stents, pacemakers, or ingested objects in children. The diversity of foreign objects complicates dense annotation, as shown in insufficient existing datasets. To address this, we propose the simple generation of synthetic data through (1) insertion of arbitrary shapes (lines, polygons, ellipses) with varying contrasts and opacities, and (2) cut-paste augmentations from a small set of semi-automatically extracted labels. These insertions are guided by anatomy labels to ensure realistic placements, such as stents appearing only in relevant vessels. Our approach enables networks to segment complex structures with minimal manually labeled data. Notably, it achieves performance comparable to fully supervised models while using 93% fewer manual annotations.

Problem

• Chest X-ray Analysis
• Foreign Object Detection
• Annotation Difficulty

Innovation

• Synthetic Image Generation
• X-ray Object Recognition
• Semi-Automated Annotation